The UN's new scientific panel on AI, established by the General Assembly in August 2025 and made up of 40 independent experts from around the world, released its first major report this week. The timing is notable: the panel's inaugural report lands just as the gap between countries that benefit from AI and those that do not is measurably growing.
The numbers tell the story. In wealthy countries, around one in four working-age people now uses AI tools regularly. In lower-income countries, the figure sits near one in seven. That gap widened over the course of 2025, with adoption in richer nations growing almost twice as fast as in poorer ones. Of the ten countries that added the most AI users last year, every single one was a high-income economy.
At the top of the rankings, the UAE has 64% of its working-age population using AI tools. Singapore is at 61%. Norway, Ireland, France, and Spain all sit above 40%. Cambodia, by contrast, records around 5%. These are not abstract statistics: they reflect real differences in what workers can do, how fast businesses can operate, and how much value an economy can extract from the technology.
The panel's report makes a point that matters for any business operating internationally: getting access to an AI tool is not the same as benefiting from it. Countries that rely on foreign-built AI systems may be able to use them, but they have no say in the standards, the safeguards, or whether the system fits local needs. The panel calls this a control problem, not just an access problem.
The language gap is one of the clearest examples of this. AI systems perform well in English and a small number of widely spoken languages. For most of the world's other languages, performance drops significantly. Nations where lower-resource languages dominate show weaker AI adoption even after accounting for income and internet access. The medical consequences are already documented: AI translation errors in less-resourced languages have produced mistakes so serious that diseases were confused with each other and medication instructions became dangerous.
This matters beyond healthcare. Any business that uses AI to communicate with customers, partners, or regulators across language boundaries faces the same underlying risk: the tool may be confidently wrong in ways that no one catches until it is too late.
The UN panel itself was established by the General Assembly in a near-consensus vote, with the United States voting against and only Paraguay joining. That political friction is worth noting. The panel has no enforcement powers: it cannot set rules, ban products, or compel companies to do anything. It is designed to produce annual science-based assessments that inform government policy discussions, similar in structure to how climate science eventually shaped climate policy. Its first report goes to a global AI governance dialogue in Geneva this week.
The underlying logic of the report is that gaps in AI capability between countries compound over time. Countries with reliable power, fast internet, digital skills, and AI-literate workforces attract investment and pull further ahead. Countries without those foundations fall further behind, and the consequences spill over: economic migration, political instability, and markets that become harder to operate in.
For a business operator, the immediate takeaway is practical. AI tools built for global use are not equally reliable across all languages, regions, or regulatory contexts. Testing matters. Assuming the tool works because it works in your language is a mistake that will surface at the worst possible time.